And having spent more than 20 years at Mentor and 21 years prior to that with semiconductor company Texas Instruments, Rhines could be excused for now putting his feet up, or least just doing exactly what he wants to do. Rhines said it had not been his intention to take another CEO position.
“I had a full agenda of other things I was doing. One of which was a project for DARPA on the availability of fully homomorphic encryption [FHE] technology,” Rhines said.
It turns out that FHE is not very available at all. And that was key to Rhines agreeing to take the top job at Cornami, which was founded by in 2011 by Gordon Campbell, Paul Master, and Fred Furtek under the name Sviral (see Wally Rhines takes CEO job at ‘sea-of-cores’ startup).
But what is FHE? It is a form of encryption that allows computation using ciphered text that generates an encrypted result that when decrypted matches the result that would have been obtained by operating on the raw data. It may sound arcane but is important because it means that outsourcing storage and computation to the cloud can be made more secure. Data centers can be allowed to process data or parts of bigger data sets without any knowledge of the data itself.
That can be as important for something as mundane as a Google Maps request for the location of a local coffee shop as it can be for emails about the travel plans of persons of significance, such as heads of state. FHE is also said not to be susceptible to quantum computing attacks, Rhines added.
DARPA is interested because it believes it will become conventional wisdom to “secure the data, not the data center,” Rhines said.
Next: Meeting Cornami
While Rhines was getting to grips with FHE, Cornami was contemplating entering the artificial intelligence/machine learning (AI/ML) market armed with simulations of their “sea-of-cores” reconfigurable systolic array processor. Rhines explains that the company had been founded under the name Sviral as a software company with the ability to compile code down to an intermediate virtual machine that targets multi-core arrays. “In the end they decided to design their own chip to take full advantage of their multicore targeting software.”
It was during a chance encounter with Cornami that Rhines was told by Paul Masters that their chip design also supports something obscure called fully homomorphic encryption. In fact, the chip can do it in real-time, Masters said. Rhines benchmarked this by saying that Cornami is looking to achieve six-orders of magnitude improvement on conventional server performance.
“I spent three or four months with the company going through the emulation data. And I was aware that no-one was predicting that this would be available any time soon, but meanwhile there are customers in areas such as financial services, eager to use the technology.
Rhines agreed to sign-on as CEO while deciding that Cornami should focus on FHE.
“Cornami’s chip design can do edge-node vision, pattern recognition, block-chain. In fact, it’s impressive at these things but it’s already a crowded market place. In FHE, the company has a free and open field; it can revolutionize the world; and cloud providers will charge a premium for the service,” reasoned Rhines.
That’s not say that Cornami won’t enjoy success in the more conventional machine learning space. The architecture scales from thousands to millions of cores and its reconfigurability allows the hardware to be more easily optimized to the algorithm. For neural networks, it can minimize memory references and be made self-pruning of redundant nodes.
Next: Going to silicon
Rhines said the next step for Cornami is to take the design to silicon. The company is targeting a 16nm FinFET process at TSMC. This is not at the bleeding-edge of 7nm or 5nm. Targeting just behind the leading-edge makes manufacturing more affordable and gives the company accurate yield data, Rhines said. He added that because the computing architecture scales in a highly linear manner it makes sense to offer an affordable single chip and scale up with multiple chips on a board.
We asked Rhines what he brings to Cornami?
“I have been through this experience before. I was with Texas Instruments when we were pitching the TMS9900 against the Intel 8086 and the Motorola 68000.” This was back in the late 1970s and early 1980s and the time of 16bit processor battles. It can be argued the Intel 8086 won when the 8088 was designed in to the IBM PC, giving rise to the x86 architecture, although the 68000 architecture went on to provide long service in embedded applications. Rhines continued: “I saw we’d already lost the microprocessor battle and decided we should go into a new field with the TMS320.”
The launch of the TMS32010 in 1983 marked the advent of the monolithic digital signal processor, a sector that was almost synonymous with TI for many years. Rhines recalls the similarities with FHE today. “We knew the defense sector was interested and we were thinking of speech processing and some other applications. But we didn’t know all the applications that would follow. Ultimately DSP got TI into mobile phones with Nokia, Ericsson and Alcatel.”
Rhines referred to the AI/ML sector as being overcrowded, but that doesn’t mean it isn’t a highly significant development.
Next: AI/ML winners
“It’s a true discontinuity. As an industry we needed to move from von Neumann computation to some form of neuromorphic computing and all sorts of innovative people have come forth. There are more than 50 startups developing chips that are AI-related. Since 2012 some $2.7 billion has been invested in AI/ML startups,” said Rhines.
“And the field is broad. Some chips are very special-purpose, some are more general. But some of the leading startups are valued at $2 billion or more on almost no sales to date.” Rhines called out Graphcore, Groq, Habana Labs and SambaNova as amongst four of those highly valued. Habana Labs’ valuation was crystallized when Intel paid $2 billion to acquire the company at the end of 2019 (see Intel pays $2 billion for AI chip firm).
“There will be multiple successful companies because I believe there are multiple applications. Also, the software does not build up around one or other hardware architecture but instead it builds up around the development platforms; PyTorch, TensorFlow and so on. But you won’t get 50 winners.”
Rhines added that the nature of market adoption would dictate that AI/ML success would come in waves. “The first wave of chips will run the same software as Nvidia and others and be able to demonstrate superiority. The second wave will come as chips achieve dominance in particular applications.”
Our conversation – conducted by Zoom – then moved on to the Covid-19 pandemic and its impact. While the social impact of the pandemic has been large, Rhines does not see it as being a catastrophic event for the semiconductor industry noting that some markets have even been stimulated by stay-at-home and work-at-home social changes. “The semiconductor market is down. It will probably be down in the second quarter. It will be a net negative for the economy in 2020. But hopefully we will see a vaccine in 2021.”
We next asked Rhines about the international trading environment. Most of Rhines’ career has coincided with an era of liberalization, lowering of tariffs, freeing up of trade, and globalization. Recently there has been a swing back to strategic self-sufficiency, protectionism and even isolation. Big trend or blip, we asked?
Next: Freedom wins out
“It is probably a bit more than a blip. I hope it will cure itself. Free trade and globalization have created millions of jobs. They have allowed specialization. It has been a wonderful era; the fuel of economic growth. With a more defensive stance you will stifle economic growth, increase the disparity between rich and poor; all negative effects. But recently politicians have been willing to pay the price.” Rhines continued: “In the US it is a very bipartisan issue. But there is a price to be paid when you bring back the low-pay jobs.”
And finally, we asked about China and its momentum towards becoming the next global superpower.
“In the long-term freedom, free communications and free trade always wins. China was opening up its markets and moving towards free trade. Recently China has become more closed, more restrictive and more focused on self-sufficiency.”
Rhines observes that over the decades other countries have competed with the US in semiconductors, such as Japan and Korea in memories and Taiwan in foundry manufacturing and China is the latest on that list.
“But the US has freedom of communication and a lack of regulation, which provide the opportunity for entrepreneurship and innovation.” It is one of the reasons that Silicon Valley started in California and continues to flourish with startups being formed there at a tremendous rate. Which neatly leads back to the matter in hand; Rhines’ hands on the reins of a US startup called Cornami.
The technical storage or access is strictly necessary for the legitimate purpose of enabling the use of a specific service explicitly requested by the subscriber or user, or for the sole purpose of carrying out the transmission of a communication over an electronic communications network.
The technical storage or access is necessary for the legitimate purpose of storing preferences that are not requested by the subscriber or user.
The technical storage or access that is used exclusively for statistical purposes.The technical storage or access that is used exclusively for anonymous statistical purposes. Without a subpoena, voluntary compliance on the part of your Internet Service Provider, or additional records from a third party, information stored or retrieved for this purpose alone cannot usually be used to identify you.
The technical storage or access is required to create user profiles to send advertising, or to track the user on a website or across several websites for similar marketing purposes.